It was resolved by itself the next day, and apparently it was related to some bugs which hopefully Databricks would crush them soon and improve the whole serverless service